Nonlinear intervention modeling with an application to time series forecasting /
| Main Author: | |
|---|---|
| Other Authors: | , , |
| Format: | Thesis Book |
| Language: | English |
| Published: |
1988.
|
| Subjects: | |
| Online Access: | ProQuest, Abstract Link to OAKTrust copy |
| Abstract: | Most time series modeling techniques have been developed under the assumption that time series are realizations of homogeneous processes. However, there are many real world situations in which time series do not satisfy that assumption because of external factors seriously affect the underlying process. If conventional time series techniques are used to model and forecast a nonhomogeneous time series, the resulting forecasts may not correspond to its future ensemble. In this research the major concern is to study nonhomogeneous time series for purposes of modeling and forecasting. Analytical and simulation techniques have been used to study time series which exhibit local nonstationary behavior. The most important task when modeling local nonstationary time series is to detect when the local disturbance occurs and to identify a more suitable model for on-line estimation and forecasting. In this research a statistical test is proposed to detect when a local nonstationary disturbance occurs and whether it affects the mean or the covariance function of the underlying time series. This test has been compared with three available tests and has been found to give the most successful performance. In this study a recursive-iterative data transformation procedure has been developed for the purpose of on-line identification and subsequent removal of nonlinear intervention effects following their detection. Local nonstationarities in time series; which are known as interventions effects in the statistical literature, can be modeled by linear transfer function. However, nonlinear transfer function models have been found to be more suitable to represent rainfall and extreme temperature intervention effects. These new models utilize on-line control variables to adapt the model to respond to temporal variabilities more effectively than the conventional intervention approach. An iterative-recursive algorithm for accurate prediction of local nonstationary time series is another contribution of this study. The algorithm is a combination of recursive parameter estimation technique, a sequential detection of local nonstationary effects, an on-line parameter updating procedure, and a recursive forecasting algorithm... |
|---|---|
| Item Description: | Typescript (photocopy). Vita. "Major subject: Industrial Engineering." |
| Physical Description: | x, 214 leaves : illustrations ; 29 cm |
| Bibliography: | Includes bibliographical references. |